noatmark-dsh-plugin

NoAtMark text hygiene as a DeepSeek Harness (dsh) plugin.

Everything-is-a-plugin: this plugin gives your dsh agent four text-hygiene tools, backed by the same deterministic engines behind noatmark.com.

Tools

| Tool | What it does | |---|---| | sanitize_text | Strip invisible/zero-width characters and flag prompt-injection + hidden-text signals. | | scan_text | Report invisible characters (with code points + positions), injection patterns, and hidden text. | | clean_format | Clean LLM formatting artifacts (blank lines, stray fences, trailing spaces) — meaning untouched. | | sanitize_csv | Escape CSV formula injection (OWASP): = + - @ prefixes get a leading quote. |

All processing is deterministic and local — no data leaves your machine.

Install

Add this plugin to your dsh Web UI. If you're running from a source checkout, use a cordis.yml patch:

- insert:
    - id: noatmark
      name: noatmark-dsh-plugin

or point at the source directly (absolute path):

- insert:
    - id: noatmark
      name: '/path/to/noatmark-dsh-plugin/src/index.ts'

Start dsh with the patch:

npx @deepseek-ai/dsh web --patch ./cordis.yml

Usage

In a dsh session, ask the agent to use a tool, e.g.:

Sanitize this pasted text before you summarize it.

Scan this file for invisible characters.

Clean the formatting of this AI output.

Escape this CSV before saving it.

Development

pnpm install
pnpm build   # tsc -> dist/

Resources

• Web tools: https://noatmark.com/
• NoAtMark SDK (npm/PyPI): noatmark-text-hygiene
• DeepSeek Harness: https://github.com/deepseek-ai/deepseek-harness
• Official docs: https://deepseek-harness.github.io/deepseek-harness/

MIT